Multi-focus image processing for visual effect generations

在这个项目中，我们研究了一种创新的方法来处理图像/视频内容，以产生各种视觉效果。在传统的方法中，图像和视频必须被分割成物体，物体被不同的视觉效果分别应用。我们的方法与传统的方法完全不同，传统的方法是使用多个不同聚焦的图像，直接生成目标图像，而不进行分割。在我们的新方法中，我们仅通过线性处理就可以生成基于对象的视觉效果。我们在这个项目中取得的具体成果如下：1)多焦点图像处理理论基础的发展。通过对多幅不同聚焦的图像进行线性处理，可以实现逐物控制各种视觉效果的渲染，如模糊、运动模糊、增强、相移等。2)预处理自动化。注册（位置，比例）需要作为预处理。所需的预处理是自动化的。3)虚拟视图生成。通过我们的方法移动前景和背景对象并生成虚拟视图。该方法被进一步应用于所谓的光场渲染。相机阵列的每个相机捕获两个不同聚焦的图像，并从中生成新的视图。由于我们的技术，实现了少量摄像机下的平滑插值。4)昆虫显微图像的图像处理。对昆虫显微图像进行处理，生成所有聚焦图像。大量显微图像用于评价该方法。同时生成了昆虫的深度图像。

In this project, we have investigated an innovative approach to image/video contents manipulation for various visual effect generation. In the conventional approach, the image and video have to be segmented into objects and objects are separately applied by various visual effects. Our approach is totally different from the conventional approach, in which we use multiple differently focused images and directly generate the target image without segmentation. In our new method, we can generate object-based visual effects onfy by linear processing. The specific achievements we have made in this project is shown below.1) Development of theoretical foundation of multi-focus image processing. Rendering of various visual effects such as blur, motion blur, enhancement, phase shift was able to be controlled object by object only by linear processing of multiple differently focused images.2) Automation of preprocessing. Registration (position, scale) is required as a preprocessing. The required preprocessing was automated.3) Virtual view generation. Foreground and background objects were shifted by our method and virtual views were generated. The method was further applied to the so called light field rendering. Each camera of the camera array captured two differently focused images and the new views were generated from them. Because of our technique, smooth interpolation under small number of camera was achieved.4) Image processing of microscopic images of insects. Microscopic images of insects were processed so that all focus image was generated. A large number of microscopic images were used for evaluation of the method. Depth image of the insects was also generated.